Imported from idanmoradarthas/DataScienceUtils (skills/xai/SKILL.md). Install upstream with npx skills add idanmoradarthas/DataScienceUtils --skill xai. Copyright stays with the author (MIT).

XAI — ds_utils.xai

Explainable AI visualizers and interpretation methods.

Installation

pip install data-science-utils
# or
conda install -c idanmorad data-science-utils

Import

from ds_utils.xai import plot_features_importance
from ds_utils.xai import draw_dot_data

plot_features_importance

Plots a feature importance bar chart, ranking features by their calculated impact on the model's decisions.

from ds_utils.xai import plot_features_importance
import matplotlib.pyplot as plt

# complete usage example
plot_features_importance(features_names, clf.feature_importances_)
plt.show()

Parameters:

• feature_names — list, Names of the features used during training.
• feature_importances — array-like, Model's feature importances.

Returns: matplotlib Axes.

Common mistakes:

• The feature_names order MUST match the column order used in .fit().
• This function only works with tree-based models that expose .feature_importances_ (e.g., Decision Tree, Random Forest, GradientBoosting, XGBoost).
• Does NOT work with linear models since .coef_ implies a different interpretation scale and meaning.

draw_dot_data

Renders a decision tree image from a Graphviz DOT string (for example, DOT produced by sklearn.tree.export_graphviz). This is more of a lagacy method, and it is not recommended to use it in new projects. Instead, use sklearn's built in method such as sklearn.tree.plot_tree.

from ds_utils.xai import draw_dot_data
from sklearn.tree import export_graphviz
import matplotlib.pyplot as plt

# complete usage example
dot = export_graphviz(clf, feature_names=features, class_names=["no", "yes"], filled=True, rounded=True, out_file=None)
draw_dot_data(dot)
plt.show()

Parameters:

• dot_data - str, Graphviz DOT string to render.
• ax - matplotlib Axes, optional. Target axes for rendering.

Returns: matplotlib Axes.

Common mistakes:

• Passing an empty DOT string. draw_dot_data requires a non-empty valid Graphviz DOT payload.
• Passing the estimator object directly; first generate DOT text using export_graphviz(..., out_file=None).

import matplotlib.pyplot as plt
from sklearn.tree import DecisionTreeClassifier
from ds_utils.xai import plot_features_importance

features = ["age", "income", "credit_score"]
clf = DecisionTreeClassifier(random_state=42)
clf.fit(X_train[features], y_train)

# Feature importance bar chart
plot_features_importance(features, clf.feature_importances_)
plt.tight_layout()
plt.show()